Search

Graph Neural Architecture Search (GNAS) has achieved superior performance on various graph-structured tasks. However, existing GNAS studies overlook the applications of GNAS in resource-constraint scenarios. This paper proposes to design a joint graph data and architecture mechanism, which identifies important sub-architectures via the valuable graph data. To search for optimal lightweight Graph Neural Networks (GNNs), we propose a Lightweight Graph Neural Architecture Search with Graph SparsIfication and Network Pruning (GASSIP) method. In particular, GASSIP comprises an operation-pruned architecture search module to enable efficient lightweight GNN search. Meanwhile, we design a novel curriculum graph data sparsification module with an architecture-aware edge-removing difficulty measurement to help select optimal sub-architectures. With the aid of two differentiable masks, we iteratively optimize these two modules to efficiently search for the optimal lightweight architecture. Extensive experiments on five benchmarks demonstrate the effectiveness of GASSIP. Particularly, our method achieves on-par or even higher node classification performance with half or fewer model parameters of searched GNNs and a sparser graph., Comment: Accepted by KDD 2024. The two first authors made equal contributions

Graph NAS has emerged as a promising approach for autonomously designing GNN architectures by leveraging the correlations between graphs and architectures. Existing methods fail to generalize under distribution shifts that are ubiquitous in real-world graph scenarios, mainly because the graph-architecture correlations they exploit might be spurious and varying across distributions. We propose to handle the distribution shifts in the graph architecture search process by discovering and exploiting the causal relationship between graphs and architectures to search for the optimal architectures that can generalize under distribution shifts. The problem remains unexplored with following challenges: how to discover the causal graph-architecture relationship that has stable predictive abilities across distributions, and how to handle distribution shifts with the discovered causal graph-architecture relationship to search the generalized graph architectures. To address these challenges, we propose Causal-aware Graph Neural Architecture Search (CARNAS), which is able to capture the causal graph-architecture relationship during the architecture search process and discover the generalized graph architecture under distribution shifts. Specifically, we propose Disentangled Causal Subgraph Identification to capture the causal subgraphs that have stable prediction abilities across distributions. Then, we propose Graph Embedding Intervention to intervene on causal subgraphs within the latent space, ensuring that these subgraphs encapsulate essential features for prediction while excluding non-causal elements. Additionally, we propose Invariant Architecture Customization to reinforce the causal invariant nature of the causal subgraphs, which are utilized to tailor generalized graph architectures. Extensive experiments demonstrate that CARNAS achieves advanced out-of-distribution generalization ability.

The exploration of outflows in protobinary systems presents a challenging yet crucial endeavour, offering valuable insights into the dynamic interplay between protostars and their evolution. In this study, we examine the morphology and dynamics of jets and outflows within the IRAS\,4A protobinary system. This analysis is based on ALMA observations of SiO(5--4), H$_2$CO(3$_{0,3}$--2$_{0,3}$), and HDCO(4$_{1,4}$--3$_{1,3}$) with a spatial resolution of $\sim$150\,au. Leveraging an astrochemical approach involving the use of diverse tracers beyond traditional ones has enabled the identification of novel features and a comprehensive understanding of the broader outflow dynamics. Our analysis reveals the presence of two jets in the redshifted emission, emanating from IRAS\,4A1 and IRAS\,4A2, respectively. Furthermore, we identify four distinct outflows in the region for the first time, with each protostar, 4A1 and 4A2, contributing to two of them. We characterise the morphology and orientation of each outflow, challenging previous suggestions of bends in their trajectories. The outflow cavities of IRAS\,4A1 exhibit extensions of 10$''$ and 13$''$ with position angles (PA) of 0$^{\circ}$ and -12$^{\circ}$, respectively, while those of IRAS\,4A2 are more extended, spanning 18$''$ and 25$''$ with PAs of 29$^{\circ}$ and 26$^{\circ}$. We propose that the misalignment of the cavities is due to a jet precession in each protostar, a notion supported by the observation that the more extended cavities of the same source exhibit lower velocities, indicating they may stem from older ejection events.

Driven by powerful image diffusion models, recent research has achieved the automatic creation of 3D objects from textual or visual guidance. By performing score distillation sampling (SDS) iteratively across different views, these methods succeed in lifting 2D generative prior to the 3D space. However, such a 2D generative image prior bakes the effect of illumination and shadow into the texture. As a result, material maps optimized by SDS inevitably involve spurious correlated components. The absence of precise material definition makes it infeasible to relight the generated assets reasonably in novel scenes, which limits their application in downstream scenarios. In contrast, humans can effortlessly circumvent this ambiguity by deducing the material of the object from its appearance and semantics. Motivated by this insight, we propose MaterialSeg3D, a 3D asset material generation framework to infer underlying material from the 2D semantic prior. Based on such a prior model, we devise a mechanism to parse material in 3D space. We maintain a UV stack, each map of which is unprojected from a specific viewpoint. After traversing all viewpoints, we fuse the stack through a weighted voting scheme and then employ region unification to ensure the coherence of the object parts. To fuel the learning of semantics prior, we collect a material dataset, named Materialized Individual Objects (MIO), which features abundant images, diverse categories, and accurate annotations. Extensive quantitative and qualitative experiments demonstrate the effectiveness of our method.

One of the main problems in astrochemistry is determining the amount of sulfur in volatiles and refractories in the interstellar medium. The detection of the main sulfur reservoirs (icy H$_2$S and atomic gas) has been challenging, and estimates are based on the reliability of models to account for the abundances of species containing less than 1% of the total sulfur. The high sensitivity of the James Webb Space Telescope provides an unprecedented opportunity to estimate the sulfur abundance through the observation of the [S I] 25.249 $\mu$m line. We used the [S III] 18.7 $\mu$m, [S IV] 10.5 $\mu$m, and [S l] 25.249 $\mu$m lines to estimate the amount of sulfur in the ionized and molecular gas along the Orion Bar. For the theoretical part, we used an upgraded version of the Meudon photodissociation region (PDR) code to model the observations. New inelastic collision rates of neutral atomic sulfur with ortho- and para- molecular hydrogen were calculated to predict the line intensities. The [S III] 18.7 $\mu$m and [S IV] 10.5 $\mu$m lines are detected over the imaged region with a shallow increase (by a factor of 4) toward the HII region. We estimate a moderate sulfur depletion, by a factor of $\sim$2, in the ionized gas. The corrugated interface between the molecular and atomic phases gives rise to several edge-on dissociation fronts we refer to as DF1, DF2, and DF3. The [S l] 25.249 $\mu$m line is only detected toward DF2 and DF3, the dissociation fronts located farthest from the HII region. The detailed modeling of DF3 using the Meudon PDR code shows that the emission of the [S l] 25.249 $\mu$m line is coming from warm ($>$ 40 K) molecular gas located at A$_{\rm V}$ $\sim$ 1$-$5 mag from the ionization front. Moreover, the intensity of the [S l] 25.249 $\mu$m line is only accounted for if we assume the presence of undepleted sulfur., Comment: 16 pages, 6 figures. Accepted for publication in Astronomy and Astrophysics

Mid-infrared emission features probe the properties of ionized gas, and hot or warm molecular gas. The Orion Bar is a frequently studied photodissociation region (PDR) containing large amounts of gas under these conditions, and was observed with the MIRI IFU aboard JWST as part of the "PDRs4All" program. The resulting IR spectroscopic images of high angular resolution (0.2") reveal a rich observational inventory of mid-IR emission lines, and spatially resolve the substructure of the PDR, with a mosaic cutting perpendicularly across the ionization front and three dissociation fronts. We extracted five spectra that represent the ionized, atomic, and molecular gas layers, and measured the most prominent gas emission lines. An initial analysis summarizes the physical conditions of the gas and the potential of these data. We identified around 100 lines, report an additional 18 lines that remain unidentified, and measured the line intensities and central wavelengths. The H I recombination lines originating from the ionized gas layer bordering the PDR, have intensity ratios that are well matched by emissivity coefficients from H recombination theory, but deviate up to 10% due contamination by He I lines. We report the observed emission lines of various ionization stages of Ne, P, S, Cl, Ar, Fe, and Ni, and show how certain line ratios vary between the five regions. We observe the pure-rotational H$_2$ lines in the vibrational ground state from 0-0 S(1) to 0-0 S(8), and in the first vibrationally excited state from 1-1 S(5) to 1-1 S(9). We derive H$_2$ excitation diagrams, and approximate the excitation with one thermal (~700 K) component representative of an average gas temperature, and one non-thermal component (~2700 K) probing the effect of UV pumping. We compare these results to an existing model for the Orion Bar PDR and highlight the differences with the observations., Comment: 26 pages, 12 figures, 3 tables. Submitted to A&A, under review (1st revision)

Large language models (LLMs) have achieved great success in many fields, and recent works have studied exploring LLMs for graph discriminative tasks such as node classification. However, the abilities of LLMs for graph generation remain unexplored in the literature. Graph generation requires the LLM to generate graphs with given properties, which has valuable real-world applications such as drug discovery, while tends to be more challenging. In this paper, we propose LLM4GraphGen to explore the ability of LLMs for graph generation with systematical task designs and extensive experiments. Specifically, we propose several tasks tailored with comprehensive experiments to address key questions regarding LLMs' understanding of different graph structure rules, their ability to capture structural type distributions, and their utilization of domain knowledge for property-based graph generation. Our evaluations demonstrate that LLMs, particularly GPT-4, exhibit preliminary abilities in graph generation tasks, including rule-based and distribution-based generation. We also observe that popular prompting methods, such as few-shot and chain-of-thought prompting, do not consistently enhance performance. Besides, LLMs show potential in generating molecules with specific properties. These findings may serve as foundations for designing good LLMs based models for graph generation and provide valuable insights and further research.

The existing graph neural architecture search (GNAS) methods heavily rely on supervised labels during the search process, failing to handle ubiquitous scenarios where supervisions are not available. In this paper, we study the problem of unsupervised graph neural architecture search, which remains unexplored in the literature. The key problem is to discover the latent graph factors that drive the formation of graph data as well as the underlying relations between the factors and the optimal neural architectures. Handling this problem is challenging given that the latent graph factors together with architectures are highly entangled due to the nature of the graph and the complexity of the neural architecture search process. To address the challenge, we propose a novel Disentangled Self-supervised Graph Neural Architecture Search (DSGAS) model, which is able to discover the optimal architectures capturing various latent graph factors in a self-supervised fashion based on unlabeled graph data. Specifically, we first design a disentangled graph super-network capable of incorporating multiple architectures with factor-wise disentanglement, which are optimized simultaneously. Then, we estimate the performance of architectures under different factors by our proposed self-supervised training with joint architecture-graph disentanglement. Finally, we propose a contrastive search with architecture augmentations to discover architectures with factor-specific expertise. Extensive experiments on 11 real-world datasets demonstrate that the proposed model is able to achieve state-of-the-art performance against several baseline methods in an unsupervised manner., Comment: NeurIPS'23

Dynamic graph neural networks (DyGNNs) currently struggle with handling distribution shifts that are inherent in dynamic graphs. Existing work on DyGNNs with out-of-distribution settings only focuses on the time domain, failing to handle cases involving distribution shifts in the spectral domain. In this paper, we discover that there exist cases with distribution shifts unobservable in the time domain while observable in the spectral domain, and propose to study distribution shifts on dynamic graphs in the spectral domain for the first time. However, this investigation poses two key challenges: i) it is non-trivial to capture different graph patterns that are driven by various frequency components entangled in the spectral domain; and ii) it remains unclear how to handle distribution shifts with the discovered spectral patterns. To address these challenges, we propose Spectral Invariant Learning for Dynamic Graphs under Distribution Shifts (SILD), which can handle distribution shifts on dynamic graphs by capturing and utilizing invariant and variant spectral patterns. Specifically, we first design a DyGNN with Fourier transform to obtain the ego-graph trajectory spectrums, allowing the mixed dynamic graph patterns to be transformed into separate frequency components. We then develop a disentangled spectrum mask to filter graph dynamics from various frequency components and discover the invariant and variant spectral patterns. Finally, we propose invariant spectral filtering, which encourages the model to rely on invariant patterns for generalization under distribution shifts. Experimental results on synthetic and real-world dynamic graph datasets demonstrate the superiority of our method for both node classification and link prediction tasks under distribution shifts., Comment: NeurIPS'23

Most low-mass stars form in stellar clusters that also contain massive stars, which are sources of far-ultraviolet (FUV) radiation. Theoretical models predict that this FUV radiation produces photo-dissociation regions (PDRs) on the surfaces of protoplanetary disks around low-mass stars, impacting planet formation within the disks. We report JWST and Atacama Large Millimetere Array observations of a FUV-irradiated protoplanetary disk in the Orion Nebula. Emission lines are detected from the PDR; modelling their kinematics and excitation allows us to constrain the physical conditions within the gas. We quantify the mass-loss rate induced by the FUV irradiation, finding it is sufficient to remove gas from the disk in less than a million years. This is rapid enough to affect giant planet formation in the disk.

Automatic Modulation Recognition (AMR) is a crucial technology in the domains of radar and communications. Traditional AMR approaches assume a closed-set scenario, where unknown samples are forcibly misclassified into known classes, leading to serious consequences for situation awareness and threat assessment. To address this issue, Automatic Modulation Open-set Recognition (AMOSR) defines two tasks as Known Class Classification (KCC) and Unknown Class Identification (UCI). However, AMOSR faces core challenges in terms of inappropriate decision boundaries and sparse feature distributions. To overcome the aforementioned challenges, we propose a Class Information guided Reconstruction (CIR) framework, which leverages reconstruction losses to distinguish known and unknown classes. To enhance distinguishability, we design Class Conditional Vectors (CCVs) to match the latent representations extracted from input samples, achieving perfect reconstruction for known samples while yielding poor results for unknown ones. We also propose a Mutual Information (MI) loss function to ensure reliable matching, with upper and lower bounds of MI derived for tractable optimization and mathematical proofs provided. The mutually beneficial CCVs and MI facilitate the CIR attaining optimal UCI performance without compromising KCC accuracy, especially in scenarios with a higher proportion of unknown classes. Additionally, a denoising module is introduced before reconstruction, enabling the CIR to achieve a significant performance improvement at low SNRs. Experimental results on simulated and measured signals validate the effectiveness and the robustness of the proposed method., Comment: 14 pages, 11 figures

Dynamic graph neural networks (DyGNNs) have demonstrated powerful predictive abilities by exploiting graph structural and temporal dynamics. However, the existing DyGNNs fail to handle distribution shifts, which naturally exist in dynamic graphs, mainly because the patterns exploited by DyGNNs may be variant with respect to labels under distribution shifts. In this paper, we propose Disentangled Intervention-based Dynamic graph Attention networks with Invariance Promotion (I-DIDA) to handle spatio-temporal distribution shifts in dynamic graphs by discovering and utilizing invariant patterns, i.e., structures and features whose predictive abilities are stable across distribution shifts. Specifically, we first propose a disentangled spatio-temporal attention network to capture the variant and invariant patterns. By utilizing the disentangled patterns, we design a spatio-temporal intervention mechanism to create multiple interventional distributions and an environment inference module to infer the latent spatio-temporal environments, and minimize the variance of predictions among these intervened distributions and environments, so that our model can make predictions based on invariant patterns with stable predictive abilities under distribution shifts. Extensive experiments demonstrate the superiority of our method over state-of-the-art baselines under distribution shifts. Our work is the first study of spatio-temporal distribution shifts in dynamic graphs, to the best of our knowledge.

Dynamic graph neural networks (DGNNs) are increasingly pervasive in exploiting spatio-temporal patterns on dynamic graphs. However, existing works fail to generalize under distribution shifts, which are common in real-world scenarios. As the generation of dynamic graphs is heavily influenced by latent environments, investigating their impacts on the out-of-distribution (OOD) generalization is critical. However, it remains unexplored with the following two major challenges: (1) How to properly model and infer the complex environments on dynamic graphs with distribution shifts? (2) How to discover invariant patterns given inferred spatio-temporal environments? To solve these challenges, we propose a novel Environment-Aware dynamic Graph LEarning (EAGLE) framework for OOD generalization by modeling complex coupled environments and exploiting spatio-temporal invariant patterns. Specifically, we first design the environment-aware EA-DGNN to model environments by multi-channel environments disentangling. Then, we propose an environment instantiation mechanism for environment diversification with inferred distributions. Finally, we discriminate spatio-temporal invariant patterns for out-of-distribution prediction by the invariant pattern recognition mechanism and perform fine-grained causal interventions node-wisely with a mixture of instantiated environment samples. Experiments on real-world and synthetic dynamic graph datasets demonstrate the superiority of our method against state-of-the-art baselines under distribution shifts. To the best of our knowledge, we are the first to study OOD generalization on dynamic graphs from the environment learning perspective., Comment: Accepted by the 37th Conference on Neural Information Processing Systems (NeurIPS 2023)

Text-attributed graphs (TAGs) are prevalent on the web and research over TAGs such as citation networks, e-commerce networks and social networks has attracted considerable attention in the web community. Recently, large language models (LLMs) have demonstrated exceptional capabilities across a wide range of tasks. However, the existing works focus on harnessing the potential of LLMs solely relying on prompts to convey graph structure information to LLMs, thus suffering from insufficient understanding of the complex structural relationships within TAGs. To address this problem, in this paper we present the Disentangled Graph-Text Learner (DGTL) model, which is able to enhance the reasoning and predicting capabilities of LLMs for TAGs. Our proposed DGTL model incorporates graph structure information through tailored disentangled graph neural network (GNN) layers, enabling LLMs to capture the intricate relationships hidden in text-attributed graphs from multiple structural factors. Furthermore, DGTL operates with frozen pre-trained LLMs, reducing computational costs and allowing much more flexibility in combining with different LLM models. Experimental evaluations demonstrate the effectiveness of the proposed DGTL model on achieving superior or comparable performance over state-of-the-art baselines. Additionally, we also demonstrate that our DGTL model can offer natural language explanations for predictions, thereby significantly enhancing model interpretability.

In an era marked by the increasing adoption of Large Language Models (LLMs) for various tasks, there is a growing focus on exploring LLMs' capabilities in handling web data, particularly graph data. Dynamic graphs, which capture temporal network evolution patterns, are ubiquitous in real-world web data. Evaluating LLMs' competence in understanding spatial-temporal information on dynamic graphs is essential for their adoption in web applications, which remains unexplored in the literature. In this paper, we bridge the gap via proposing to evaluate LLMs' spatial-temporal understanding abilities on dynamic graphs, to the best of our knowledge, for the first time. Specifically, we propose the LLM4DyG benchmark, which includes nine specially designed tasks considering the capability evaluation of LLMs from both temporal and spatial dimensions. Then, we conduct extensive experiments to analyze the impacts of different data generators, data statistics, prompting techniques, and LLMs on the model performance. Finally, we propose Disentangled Spatial-Temporal Thoughts (DST2) for LLMs on dynamic graphs to enhance LLMs' spatial-temporal understanding abilities. Our main observations are: 1) LLMs have preliminary spatial-temporal understanding abilities on dynamic graphs, 2) Dynamic graph tasks show increasing difficulties for LLMs as the graph size and density increase, while not sensitive to the time span and data generation mechanism, 3) the proposed DST2 prompting method can help to improve LLMs' spatial-temporal understanding abilities on dynamic graphs for most tasks. The data and codes are publicly available at Github., Comment: Accepted by ACM KDD'24

(Abridged) We investigate the impact of radiative feedback from massive stars on their natal cloud and focus on the transition from the HII region to the atomic PDR (crossing the ionisation front (IF)), and the subsequent transition to the molecular PDR (crossing the dissociation front (DF)). We use high-resolution near-IR integral field spectroscopic data from NIRSpec on JWST to observe the Orion Bar PDR as part of the PDRs4All JWST Early Release Science Program. The NIRSpec data reveal a forest of lines including, but not limited to, HeI, HI, and CI recombination lines, ionic lines, OI and NI fluorescence lines, Aromatic Infrared Bands (AIBs including aromatic CH, aliphatic CH, and their CD counterparts), CO2 ice, pure rotational and ro-vibrational lines from H2, and ro-vibrational lines HD, CO, and CH+, most of them detected for the first time towards a PDR. Their spatial distribution resolves the H and He ionisation structure in the Huygens region, gives insight into the geometry of the Bar, and confirms the large-scale stratification of PDRs. We observe numerous smaller scale structures whose typical size decreases with distance from Ori C and IR lines from CI, if solely arising from radiative recombination and cascade, reveal very high gas temperatures consistent with the hot irradiated surface of small-scale dense clumps deep inside the PDR. The H2 lines reveal multiple, prominent filaments which exhibit different characteristics. This leaves the impression of a "terraced" transition from the predominantly atomic surface region to the CO-rich molecular zone deeper in. This study showcases the discovery space created by JWST to further our understanding of the impact radiation from young stars has on their natal molecular cloud and proto-planetary disk, which touches on star- and planet formation as well as galaxy evolution., Comment: 52 pages, 30 figures, submitted to A&A

The JWST has captured the most detailed and sharpest infrared images ever taken of the inner region of the Orion Nebula, the nearest massive star formation region, and a prototypical highly irradiated dense photo-dissociation region (PDR). We investigate the fundamental interaction of far-ultraviolet photons with molecular clouds. The transitions across the ionization front (IF), dissociation front (DF), and the molecular cloud are studied at high-angular resolution. These transitions are relevant to understanding the effects of radiative feedback from massive stars and the dominant physical and chemical processes that lead to the IR emission that JWST will detect in many Galactic and extragalactic environments. Due to the proximity of the Orion Nebula and the unprecedented angular resolution of JWST, these data reveal that the molecular cloud borders are hyper structured at small angular scales of 0.1-1" (0.0002-0.002 pc or 40-400 au at 414 pc). A diverse set of features are observed such as ridges, waves, globules and photoevaporated protoplanetary disks. At the PDR atomic to molecular transition, several bright features are detected that are associated with the highly irradiated surroundings of the dense molecular condensations and embedded young star. Toward the Orion Bar PDR, a highly sculpted interface is detected with sharp edges and density increases near the IF and DF. This was predicted by previous modeling studies, but the fronts were unresolved in most tracers. A complex, structured, and folded DF surface was traced by the H2 lines. This dataset was used to revisit the commonly adopted 2D PDR structure of the Orion Bar. JWST provides us with a complete view of the PDR, all the way from the PDR edge to the substructured dense region, and this allowed us to determine, in detail, where the emission of the atomic and molecular lines, aromatic bands, and dust originate.

(Abridged) Mid-infrared observations of photodissociation regions (PDRs) are dominated by strong emission features called aromatic infrared bands (AIBs). The most prominent AIBs are found at 3.3, 6.2, 7.7, 8.6, and 11.2 $\mu$m. The most sensitive, highest-resolution infrared spectral imaging data ever taken of the prototypical PDR, the Orion Bar, have been captured by JWST. We provide an inventory of the AIBs found in the Orion Bar, along with mid-IR template spectra from five distinct regions in the Bar: the molecular PDR, the atomic PDR, and the HII region. We use JWST NIRSpec IFU and MIRI MRS observations of the Orion Bar from the JWST Early Release Science Program, PDRs4All (ID: 1288). We extract five template spectra to represent the morphology and environment of the Orion Bar PDR. The superb sensitivity and the spectral and spatial resolution of these JWST observations reveal many details of the AIB emission and enable an improved characterization of their detailed profile shapes and sub-components. While the spectra are dominated by the well-known AIBs at 3.3, 6.2, 7.7, 8.6, 11.2, and 12.7 $\mu$m, a wealth of weaker features and sub-components are present. We report trends in the widths and relative strengths of AIBs across the five template spectra. These trends yield valuable insight into the photochemical evolution of PAHs, such as the evolution responsible for the shift of 11.2 $\mu$m AIB emission from class B$_{11.2}$ in the molecular PDR to class A$_{11.2}$ in the PDR surface layers. This photochemical evolution is driven by the increased importance of FUV processing in the PDR surface layers, resulting in a "weeding out" of the weakest links of the PAH family in these layers. For now, these JWST observations are consistent with a model in which the underlying PAH family is composed of a few species: the so-called 'grandPAHs'., Comment: 25 pages, 10 figures, to appear in A&A

Large models have emerged as the most recent groundbreaking achievements in artificial intelligence, and particularly machine learning. However, when it comes to graphs, large models have not achieved the same level of success as in other fields, such as natural language processing and computer vision. In order to promote applying large models for graphs forward, we present a perspective paper to discuss the challenges and opportunities associated with developing large graph models. First, we discuss the desired characteristics of large graph models. Then, we present detailed discussions from three key perspectives: representation basis, graph data, and graph models. In each category, we provide a brief overview of recent advances and highlight the remaining challenges together with our visions. Finally, we discuss valuable applications of large graph models. We believe this perspective can encourage further investigations into large graph models, ultimately pushing us one step closer towards artificial general intelligence (AGI). We are the first to comprehensively study large graph models, to the best of our knowledge., Comment: Accepted by NeurIPS 2023 New Frontiers in Graph Learning Workshop. Comments are welcome

Graph Neural Networks (GNNs) obtain tremendous success in modeling relational data. Still, they are prone to adversarial attacks, which are massive threats to applying GNNs to risk-sensitive domains. Existing defensive methods neither guarantee performance facing new data/tasks or adversarial attacks nor provide insights to understand GNN robustness from an architectural perspective. Neural Architecture Search (NAS) has the potential to solve this problem by automating GNN architecture designs. Nevertheless, current graph NAS approaches lack robust design and are vulnerable to adversarial attacks. To tackle these challenges, we propose a novel Robust Neural Architecture search framework for GNNs (G-RNA). Specifically, we design a robust search space for the message-passing mechanism by adding graph structure mask operations into the search space, which comprises various defensive operation candidates and allows us to search for defensive GNNs. Furthermore, we define a robustness metric to guide the search procedure, which helps to filter robust architectures. In this way, G-RNA helps understand GNN robustness from an architectural perspective and effectively searches for optimal adversarial robust GNNs. Extensive experimental results on benchmark datasets show that G-RNA significantly outperforms manually designed robust GNNs and vanilla graph NAS baselines by 12.1% to 23.4% under adversarial attacks., Comment: Accepted as a conference paper at CVPR 2023

The unprecedented urbanization in China has dramatically changed the urban spatial structure of cities. With the proliferation of individual-level geospatial big data, previous studies have widely used the network abstraction model to reveal the underlying urban spatial structure. However, the construction of network abstraction models primarily focuses on the topology of the road network without considering individual travel flows along with the road networks. Individual travel flows reflect the urban dynamics, which can further help understand the underlying spatial structure. This study therefore aims to reveal the intra-urban spatial structure by integrating the road network abstraction model and individual travel flows. To achieve this goal, we 1) quantify the spatial interaction relatedness of road segments based on the Word2Vec model using large volumes of taxi trip data, then 2) characterize the road abstraction network model according to the identified spatial interaction relatedness, and 3) implement a community detection algorithm to reveal sub-regions of a city. Our results reveal three levels of hierarchical spatial structures in the Wuhan metropolitan area. This study provides a data-driven approach to the investigation of urban spatial structure via identifying traffic interaction patterns on the road network, offering insights to urban planning practice and transportation management.